Jump to content

Recommended Posts

Posted

Just a thought

 

Any chance of a duel DCSW, one optimised for rotary wing & one optimised for fixed wing ? It is getting a little long in the tooth arriveing at a target area in the KA50, to find it has allready been cleared by A10's or other. Not being able to use tree's for cover in an attack hellicopter is a bit of an immersion killer IMHO. I would immagine there would not be many real time KA50 pilots alive if this was the case, perhaps this is why the KA50 was decomisioned lol ?

Posted

It's sounds more like a bad mission design.

 

Regarding the trees I wouldn't hold my breath on that, at least not for the Georgian map

If you had a bottom end graphics card, and needed to turn the graphics down to play,

and you wanted to play online ( meaning you can't see all the trees that others with better cards can see),

and trees had a collision model,

which option would you rather force on everyone:

1/ Force everyone online to the lowest graphic level. Everyone has the same object to dodge, poor carded players can play, but graphics are 'poor' for all.

2/ Force graphics to some other level (high) for all. Again everyone gets the same (higher) number of trees to avoid, but some people can't play online because they have a slideshow.

3/ Put the collision models there for all trees at maximum graphic level, but not render some on low graphics. Everyone can play with decent FPS, but players with graphics turned down get to play Russian roulette with invisible tree collision models when close to the ground (for example while flying a helicopter, or maybe doing pop up ground attack work..)

4/ let everyone pick their own graphic level, only have collision models for visible trees, and have people on low graphic setting fly where people on high graphic settings can't..

 

5/ Make trees a 'graphic' element only - don't put collision models on trees till everyone's video card can render all the available trees comfortably. That way it's 'fair' to everyone online regardless of the money that's availalbe for their video card, but people complain about 'ghost' trees.

Posted
Regarding the trees I wouldn't hold my breath on that, at least not for the Georgian map

 

Modeling collision detection for heavily forested area is not necessary problematic in regards to performance. It depends on distribution of trees and into how much clusters they can be grouped. The worst case is when there are lots of clusters, which contain only one tree. Typically, such one tree clusters are usually ignored (no collision model for them, and their rendering can be turned off) in games for performance reasons.

Wir sehen uns in Walhalla.

Posted
Modelling collision detection for heavily forested area is not necessary problematic in regards to performance. It depends on distribution of trees and into how much clusters they can be grouped. The worst case is when there are lots of clusters, which contain only one tree. Typically, such one tree clusters are usually ignored (no collision model for them, and their rendering can be turned off) in games for performance reasons.

 

I might be misunderstanding what you mean :) but if you're thinking that clusters of trees could have a single collision box for the whole cluster under the current engine, how do ground units get through those clusters (forests) ? How do DCS.CA players move through forests ? How does AI detection through forests vs Human detection through forests work ?

& again -

What happens when I have my graphics turned down so low I don't have some of the clusters you have ? Or would you never remove clusters, only thin them out graphically (?) - but then why can't ground units move through between the sparse trees ?

Cheers.

Posted

Clearly a not very well designed mission. Kamov can use FARPS unlike any other DCS aircraft, so use them. Let those A10C take off a good 150/200km from target points, and place a FARP for you at 20/25 km. Make also sure that you have a proper target distribution among the units. Choppers and FW are living in different worlds, so they can perform very specific missions.

 

A few typical scenario for your missions with A10s:

- KA-50 assigned to clear the air defenses to let the A10C deal with the tank convoys.

 

- Overcast cloud layer at 800/1000 m. No issue for Kamov but a bit dangerous for A10C, even more with some air defense in place. Get the lat/lon coordinates of fixed targets and send them through voice VHF to A10C orbiting above the cloud layer so they can release their GPS guided GBUs. Works very well.

 

- Urban operations with crowd control, snipers and RPG around street corners to be dealt with by Kamovs. Preventing armored re inforcements from entering the city for A10s.

 

...

Posted

Yeah, bad mission design. In one of my missions called "Taking Krymsk", the A-10Cs are held off until they manage to perform SEAD on enemy SAMs around Krymsk. By that time, the Ka-50s in the missions have arrived at the FOB ready to refuel and rearm for some action. ;)

 

I'm thinking of upgrading it to current version.

Nice plane on that gun...

OS764 P930@4 MBUD3R M6GB G5870 SSDX25 CAntec1200 HTMHW

Posted
I might be misunderstanding what you mean :) but if you're thinking that clusters of trees could have a single collision box for the whole cluster under the current engine, how do ground units get through those clusters (forests) ? How do DCS.CA players move through forests ? How does AI detection through forests vs Human detection through forests work ?

& again -

What happens when I have my graphics turned down so low I don't have some of the clusters you have ? Or would you never remove clusters, only thin them out graphically (?) - but then why can't ground units move through between the sparse trees ?

 

At first, collision detection and graphical representation of trees are 2 different problems. Difference lies mainly in fact, that graphical representation (rendering) can be tied to so called player bubble (i.e. you can render trees only around X km from player/camera and so on), and collision detection can't (because there can be lots of interacting objects across whole map and you need to always check collision for them all). The main problem of collision detection is quickly checking tree presence at arbitrary location. "Quickly" and "arbitrary" basically means keeping whole required data in memory.

 

Collision detection with terrain is usually done by simply checking the height above ground. Terrain elevation is usually read from height map, which basically is 2D array, so reading height of particular point is very fast. One can think, that having second height (or tree) map, where trees are marked, could be used as solution for collision detection, however this is only possible on small maps or with very crudely formed forests as more precise tree placement would require tree map with too high resolution, which would consume too much memory. So we need some other solution, something compact, but fast enough. Quadtree data structure could be that something.

 

Quadtree is data structure, in which each node has no more than 4 child nodes. Basically, you look at whole map as single rectangle (root node), then divide it into 4 subrectangles (nodes), then divide each subrectangle into another 4 subrectangles and so on, until subrectangle size decreases to 3m x 3m or so (chosen cell size for tree). After quadtree is constructed, you do the "clean up": nodes, whose all child nodes do not contain trees or all of them contain trees, are either marked "free" or "forest' accordingly, and their all child nodes are removed. If you want to know if there is a tree (object) at X,Y, you walk along quadtree: you begin at root, then choose child node (rectangle into which X,Y falls in), then choose its child node and so on until you reach node marked with "forest", "free" or "tree".

 

How deep the quadtree is (how much the search takes), and how much memory data needs, depends on how clustered the trees (objects) are. If they are heavily clustered (lots of big forests) and distribution of clusters is not uniform, performance can be pretty high and memory usage low. I hope it clears a bit what is behind "clusters" I have used in previous post.

Wir sehen uns in Walhalla.

Posted (edited)
I might be misunderstanding what you mean :) but if you're thinking that clusters of trees could have a single collision box for the whole cluster under the current engine, how do ground units get through those clusters (forests) ? How do DCS.CA players move through forests ?

Selective collision! This is such a non-issue, I can't even express it. I have seen invulnerable aircraft flying in and out of the ocean with my own eyes, so the functionality already exists in the engine. Planes should blow up when they hit trees, but bullets and tanks do not.

 

How does AI detection through forests vs Human detection through forests work ?
The AI looks through mountains (for shame), so it's not going to make the problem any worse.

 

What happens when I have my graphics turned down so low I don't have some of the clusters you have ? Or would you never remove clusters, only thin them out graphically (?) - but then why can't ground units move through between the sparse trees ?

I play below minimum requirements on a 32 bit system, and draw my trees out to 3000m. The game should absolutely force player's systems to draw trees within about 500m for the purposes of collision. And at ranges where the trees are not drawn, make ground units in forests harder to see. Players with computers this bad are already disadvantaged heavily by low FPS and resolution, so don't lose sleep over giving them some small perks.

 

Why do I get the feeling that not much thought has gone into this?

 

At first, collision detection and graphical representation of trees are 2 different problems. Difference lies mainly in fact, that graphical representation (rendering) can be tied to so called player bubble (i.e. you can render trees only around X km from player/camera and so on), and collision detection can't (because there can be lots of interacting objects across whole map and you need to always check collision for them all). The main problem of collision detection is quickly checking tree presence at arbitrary location. "Quickly" and "arbitrary" basically means keeping whole required data in memory.

I accept that tree collision is unlikely to happen in the near term unless there are significant half-solutions and workarounds. I believe that trees need to collide aircraft immediately. Now, if only player aircraft collide with trees, why can't the engine ONLY track the collision model of forests when inside the render bubble? That is, track the mutual position of a handful of tree clusters and only one plane (client-side) when inside the render distance of the trees themselves?

 

Collision detection with terrain is usually done by simply checking the height above ground. Terrain elevation is usually read from height map, which basically is 2D array, so reading height of particular point is very fast. One can think, that having second height (or tree) map, where trees are marked, could be used as solution for collision detection, however this is only possible on small maps or with very crudely formed forests as more precise tree placement would require tree map with too high resolution, which would consume too much memory. So we need some other solution, something compact, but fast enough. Quadtree data structure could be that something.

 

Once again, since most trees in the game are the same height, why not simply take an average height value and apply it to land occupied by dense forests? If we fly through the very tip of a tree, who cares, so long as a solid collision with the average-height tree will be fatal?

Edited by maturin
Posted
Once again, since most trees in the game are the same height, why not simply take an average height value and apply it to land occupied by dense forests? If we fly through the very tip of a tree, who cares, so long as a solid collision with the average-height tree will be fatal?

 

Because then you'll have ground units hovering at the rooftops. Tarzan is cool, but MBT Tarzan isn't. :P

[sIGPIC][/sIGPIC]

Daniel "EtherealN" Agorander | Даниэль "эфирныйн" Агорандер

Intel i7 2600K @ 4.4GHz, ASUS Sabertooth P67, 8GB Corsair Vengeance @ 1600MHz, ASUS GTX 560Ti DirectCU II 1GB, Samsung 830series 512GB SSD, Corsair AX850w, two BENQ screens and TM HOTAS Warthog

DCS: A-10C Warthog FAQ | DCS: P-51D FAQ | Remember to read the Forum Rules |

|
| Life of a Game Tester
Posted

ZaltysZ :

 

I did misunderstand you - I thought you were talking about approaches that are possible in the current game engine.

 

Maturin :

 

Selective collision is something we should be trying to remove from the game, not add more of :), and while your idea of forcing the rendering everything for high scenes inside 500m might go some way towards answering the "how do I allow for different graphics setting" question, my personal opinion is that it would be a very ugly, immersion killing solution (more immersion killing than non-collidable trees) - that would be some NASTY pop-up...

 

The response to most of your other questions (bubble collision calcs etc.) would be - that's not how the engine works.

 

I think you summed it up yourself :

I accept that tree collision is unlikely to happen in the near term unless there are significant half-solutions and workarounds.

 

I'm hopeful that the new DCS game engine currently under development (Edge) will address these issues in an effective way that doesn't require hacks...

Cheers.

Posted (edited)
Because then you'll have ground units hovering at the rooftops. Tarzan is cool, but MBT Tarzan isn't. :P

We were talking about aircraft collision. Aircraft constantly track distance to the ground layers, while ground vehicles stick to it.

 

Theoretically you could add another ground collision layer 5m (or however tall your average tree is) to the collision height of forested ground. Vehicles would slide underneath it on the main ground layer, though you would have to set their projectiles to fire through it (or not, depending on AI vision decisions).

 

Selective collision is something we should be trying to remove from the game, not add more of smile.gif, and while your idea of forcing the rendering everything for high scenes inside 500m might go some way towards answering the "how do I allow for different graphics setting" question, my personal opinion is that it would be a very ugly, immersion killing solution (more immersion killing than non-collidable trees) - that would be some NASTY pop-up...
A minimum of 500m, for the tiny percentage of users who play without trees at all. Given the astounding hardware elitism on these forums, there would be no outcry about some people have to suffer from pop-up. Because tree collision is a gameplay issue, not an immersion issue, especially if it can be made to prevent tanks shooting through forests and ruining CA.

 

The response to most of your other questions (bubble collision calcs etc.) would be - that's not how the engine works.
As simple as that, eh? You'll excuse me for suspecting that engine limitations are a scapegoat for lack of trying, here. But I do learn more every day about how crappy this engine is and how little will there is to address its issues. Edited by maturin
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...