Game physics

[Pyronick 21]

Yes this game truly does need a physics engine and then dedicate an entire core of a quad-core CPU to handle it. That would make my Fing... YEAR.

I think that precious CPU threads should be used for AI, asset streaming and other game logics.

And that gameplay physics should run on GPU stream processors, they can proces floating point operations hundreds of times faster, with more precision.

On top of that GPUs are becoming so fast that physics engines would hardly to not tax the framerate in anyway.

The new AMD RV870 has 2.7 TFLOPS, which is equivalent to a BlueGene supercomputer of four years ago.

And there will be a dual-GPU setup called R800, which has 5.4 TFLOPS.

It is pure overkill, as the ROPs and TMUs are usually operating on full speed all while the stream processors are almost idling.

Edited September 13, 2009 by SgtH3nry3

[Kalikorz 10]

In all Seriousness

Separating pieces on vehicles and aircraft would solve these collision problems and be realistic. If your plane slams into another plane the wings fly of the tail flies over the top and the middle scrunches and falls all while a massive fuel air explosion is taking place. The engine can handle an on average 10 objects being thrown from an impact between two things. I would love to see heavy logs flying through the air after an artillery strike also. Just find a comfortable limit for the engine let it fly and increase the friction so that stuff actually comes to a stop removing it's pressure on the physics engine after less than a minute under normal circumstances. This could also be applied with the earlier tracks-off-the-tank scenario. You may believe this would be too much for the engine but I am begging you TRY IT. even one or two flying pieces would be better than the shrapnel-less world we are in now.

OTHER NOTES

- Watching an AI tank jump and walk at a 60 degree angle is killing me.

- Please first shot accuracy is not realistic with iron sighted weapons and machine guns. It's barely realistic with ACOG's. Usually a soldier cold-firing a weapon drops one round off and walks onto the target from there. A sniper who has not already been firing his weapon at range should not have instant accuracy (think SWAT 3). Semi- auto AI would have quickly increasing accuracy with ok precision starting at moderatly inaccurate. Burst AI would have moderate starting accuracy with better precision getting better at an average rate and more space between bursts than the semi has between shots. Full- auto AI would have Bad initial accuracy with slowly increasing accuracy and moderate precision. Even more so an AI must be able to see the shot land in order to correct onto his target. We have all had the experience of firing at someone on a ridgeline and not being able to hit him forever because we can't tell which way we are missing.

Argue or tweak as you must but the idea of increasing accuracy over time and not starting with perfect ESPECIALLY without unobstructed line of site is the point.

- Planes are nearly unusable with out increased visability and some kind of targeting system.

My gripes are my own take them or leave them just don't hold them against me

Edited September 16, 2009 by Kalikorz

[Rapier 10]

Flight physics+ flight model are terrible.

I found out why I can't speed up enough not to collide with the ground many times. Because, I'm trying to fly like in real life.

Why are the plane brakes auto-engage during nose down?

I don't want to be pushing the accelerator all the time, come on!

If i want to pick up speed from not stalling, it's obvious that most pilots will nose down the plane to descent to gain speed, even kindergarten kids are aware of this. Why even implement such a thing?

If this game try to simulate such a small things like sun flare and fuzzy view while running and so on, why are the big details like physics and flight model left for arcade physics? It's not even arcade physics though it's worse, because the game is working against me with ideas like this.

Also try this.

If you turn, you'll loose more speed than climb. I can actually accelerate faster by straight nose up to the sky than trying to turn around. This is not only bad, it's stupid.

IN real life, every plane has it's rate of climb and rate of turn, which depends on it's flight model, engine power weight, etc.

Each every one of them has its range of speed that is given which would give the best speed vs turn ratio.

But I have yet to see a plane that turns faster if the speed is faster , especially if it's outside the turn/speed ratio. It's straight against Newtonian physics.

Edited September 25, 2009 by Rapier

[nomdeplume 0]

Why are the plane brakes auto-engage during nose down?

Are you using analog throttle control or the standard controls? I've not used the analog throttle so I'm not sure how it behaves (the implementation just sounded too weird for me to bother trying it). So, probably disregard this post if you are using the analog control.

With the standard "digital" controls, it's actually more of a cruise control than throttle control. I don't know if you've realised this or not, but you didn't mention it in your post so maybe not. It took me a while to realise what the plane was doing but now that I understand it it's easier to fly. It's not perfect, but at least I understand what the heck it's doing.

Anyway, the jist is that while you hold the accelerate or brake keys, you apply full thrust or full airbrake (obviously). But here's the kicker: when you let go of the key, your current speed at the moment you let go is recorded, and the plane tries to maintain that until the next time you touch one of the throttle controls.

Let's say you take off and level out at 400 kph. If you do a maneuver that causes you to lose speed, e.g. climbing or turning tightly, the throttle will open up and try to bring you back to 400. Likewise if you dive you'll pick up speed, but the game will apply the brakes to try to keep you at the target speed.

I haven't experimented to see if your throttle is more powerful than the auto-throttle when it tries to get you back up to speed, but my guess is that there's no difference. The upshot of this assumption is that there is no point in touching either of the throttle controls unless you want to change your "target speed".

The other important thing to note is that if you do a maneuver which kills your airspeed and you hit the throttle key for a bit (e.g. to avoid a stall) then release it, you've just re-set your target speed - probably to well below your original target speed. If you then nose down to try to pick up some extra speed, the plane will try to prevent you from exceeding your newly-set target speed by applying the brakes.

Yes, the flight model and controls are very simplified, but I think that's kind of the point. The game isn't a flight sim, and it's trying to make it so that everyone who plays the game can learn to fly a plane or helo, drive a tank, etc. and contribute to the battlefield in the manner in which those weapons platforms tend to, without having to spend dozens of hours learning the details of flight capabilities and avionics. This grates with some people who think you shouldn't be allowed to fly the aircraft until you've practiced for 100 hours offline... but I don't think this is the game for them.

You do still need to practice to get a feel for it, and if you're a flight sim fan there's quite a bit of adjustment from how you know aircraft fly to how they fly in the game. But again - it's a flight game, not a flight sim.

[Rapier 10]

You do still need to practice to get a feel for it, and if you're a flight sim fan there's quite a bit of adjustment from how you know aircraft fly to how they fly in the game. But again - it's a flight game, not a flight sim.

I'm flying a keyboard+mouse control. I understand the point of "setting a cruise speed" but at the same time, (and you can try this), there is a certain point at a certain speed the plane will keep the cruising speed, but under this speed (it's usually around 400km/h) the plane will not hold the set speed but slows down and eventually falls out of the sky. And even if you set this cruising speed, go try it, do a turn and when you get out of that turn and return to level wing straight-notice your plane no longer holds that cruising speed and it is neither accelerating to regain that speed.

Of course, you can say that "oh well" that speed isn't fast enough to keep the plane in the air, but then I say "you just said the game isn't a fight sim and not to expect a plane simulation here". Not to mention these planes (Harrier, SU25 etc.) are well capable of flying (and not falling out of the sky) under 300km/h, and as you say that the plane tries to keep the cruising speed I have set with my "Q" key, how do you explain that it won't really hold that speed in a level flight, but instead it rapidly drops and eventually I crash and burn unless I continuously accelerate by holding down the Q key?

What you're implying, it worked in Operation Flashpoint perfectly. It was actually enjoyable to fly around with the jets, even with the arcade controls. But not in Arma or ArmaII.

Look, it's either not a flight sim and let's have me those aids so I can play, or it is a flight sim and have the plane behave like it supposed to.

It can't be both and neither of them at the same time, because then it's a giant mess.

Edited October 1, 2009 by Rapier

[draeath 10]

- Please first shot accuracy is not realistic with iron sighted weapons and machine guns. It's barely realistic with ACOG's. Usually a soldier cold-firing a weapon drops one round off and walks onto the target from there. A sniper who has not already been firing his weapon at range should not have instant accuracy (think SWAT 3). Semi- auto AI would have quickly increasing accuracy with ok precision starting at moderatly inaccurate. Burst AI would have moderate starting accuracy with better precision getting better at an average rate and more space between bursts than the semi has between shots. Full- auto AI would have Bad initial accuracy with slowly increasing accuracy and moderate precision. Even more so an AI must be able to see the shot land in order to correct onto his target. We have all had the experience of firing at someone on a ridgeline and not being able to hit him forever because we can't tell which way we are missing.

Argue or tweak as you must but the idea of increasing accuracy over time and not starting with perfect ESPECIALLY without unobstructed line of site is the point.

I have to ask. Have you actually fired guns before?

Especially in regards to a sniper. Snipers tend to use hand picked or hand-crafted/filled ammunition, and most certainly have sighted their scope for a particular distance. They also know how to use their scope to estimate distance extremely well. Put those together, and they WILL hit what they shoot at, the first shot.

Heck, even me - a beginner - can sight my scope one day, come back the next, and hit what I'm trying to hit.

[charon productions 10]

Ok i don`t wanna critize the game physics, but rather find a work-around for specific situations. What config parameter could i possibly use to make objects less bouncing around ?

I have used 8000 kg weighting on a small object and still it bounces around half Chernarus when dropped from a plane.

I am considering reducing the velocity in regular intervals by script, but maybe there is something that i have not considered?

EDIT: That braking script does the job quite well, just tested it.

Edited October 7, 2009 by Charon Productions

[BigJonnyC 10]

Earlier I was testing the BMP on the mobility course and after clearing a ramp my vehicle flew backwards in the air for about 100-120 feet and landed next to a none-the-wiser grazing cow.

I laughed and cursed loudly.

[JuggernautOfWar 1]

Earlier I was testing the BMP on the mobility course and after clearing a ramp my vehicle flew backwards in the air for about 100-120 feet and landed next to a none-the-wiser grazing cow.

I laughed and cursed loudly.

If only if only you were recording... :D

[Psychopathus 10]

I believe it's been asked before (I tried searching but it doesn't work well...) so: are projectiles' trajectory affected by wind and/or weather conditions?

I've been playing for a while and the ballistics seem to be those from good-old OFP...

Edited November 6, 2009 by Psychopathus

[Psychopathus 10]

Earlier I was testing the BMP on the mobility course and after clearing a ramp my vehicle flew backwards in the air for about 100-120 feet and landed next to a none-the-wiser grazing cow.

I laughed and cursed loudly.

Yeah, funny Arma physics!

[McArcher 0]

even Half-Life 2 has better physics than Arma 2 =(

i want good physics and damaging model at least.

even in carmageddon you were able to break parts of your car and leave them lying on the ground !!! what we see now..... i can't even see holes in armor after subcaliber shot in it...

[Hvymtal 1251]

I think driving speeds can definately be improved, the fastest you can go (in sp at least) is 120 KPH, which is pretty slow. btw is there any mods out there that tweak this?

[Pyronick 21]

The whole driving model should be changed. In this game, most land-based vehicles drive either too easy or too difficult.

Instead of increasing the turn radius linear with speed, there should be genuine under- and oversteer in the game. Nothing really complicated though as it still should be a milsim, not a driving sim.

Braking distances should be bigger aswell, and proper acceleration should be in the game too.

[Dampfnudel 10]

Hi,

good suggestions here. I would like to add also one. Its about the drive physics. It feels like its even worse than in arma1.

Sorry but even some psx and super nintendo arcade games that are some hell years old got better drive physics. It dont need to be like in GTAIV ( would be awesome ) but at least a bit more drift characteristics. Arma/Arma2 are great games but physics should be really taken care of.

[McArcher 0]

I think driving speeds can definately be improved, the fastest you can go (in sp at least) is 120 KPH, which is pretty slow. btw is there any mods out there that tweak this?

all the speed accelerations of different vehicles are wrong in the game.

[ZeosPantera 0]

even Half-Life 2 has better physics than Arma 2 =(

i want good physics and damaging model at least.

even in carmageddon you were able to break parts of your car and leave them lying on the ground !!! what we see now..... i can't even see holes in armor after subcaliber shot in it...

The whole driving model should be changed. In this game, most land-based vehicles drive either too easy or too difficult.

Instead of increasing the turn radius linear with speed, there should be genuine under- and oversteer in the game. Nothing really complicated though as it still should be a milsim, not a driving sim.

Braking distances should be bigger aswell, and proper acceleration should be in the game too.

all the speed accelerations of different vehicles are wrong in the game.

There are no physics in Arma. There are animations and fake physics.

[Li0n 10]

Please do the animation of running/walking with pistol more realistic. For now it is looks like unit swallowed a stick.

[NeMeSiS 11]

There are no physics in Arma. There are animations and fake physics.

Fake physics? :rolleyes:

[Flash Thunder 10]

theres very simple physics in the game like collision detection, g' forces :)

Look at Dragon Rising for example, the game has to transfer alot more data to each client in the outdated retarded (PVP architecture) so usually games online have excessive pings.

Also codemasters probably assumed ragdoll on dead corpses wouldn't take alot of time from the processor to calculate, so in the game you see lower frame rates (depending on CPU) and thats why the game even though was a console port to PC which should've meant more optimization, meant the games graphics had to be scaled down, so the GPU/CPU could have more time with the physic components of DR.

Physics are nice but they come at a large price. Performance and graphical fidelity. :(

[Chunk3ym4n 10]

theres very simple physics in the game like collision detection, g' forces :)

Look at Dragon Rising for example, the game has to transfer alot more data to each client in the outdated retarded (PVP architecture) so usually games online have excessive pings.

Also codemasters probably assumed ragdoll on dead corpses wouldn't take alot of time from the processor to calculate, so in the game you see lower frame rates (depending on CPU) and thats why the game even though was a console port to PC which should've meant more optimization, meant the games graphics had to be scaled down, so the GPU/CPU could have more time with the physic components of DR.

Physics are nice but they come at a large price. Performance and graphical fidelity. :(

The ragdolls can be handled clientside to disregard the sync between server and client. The problem is syncing where the body is so you can take the gear out.

[Muzzwezz 10]

The ragdolls can be handled clientside to disregard the sync between server and client. The problem is syncing where the body is so you can take the gear out.

I think the usual method is for the client and server to sync a agreed "general trajectory" [say its a body being blown up] or "flight path"

then the client will do all the small actions e.g. the "falling" of the body, that will fall along the synced flight path,

hence why games often bodies are in slightly different positions on different clients

[MulleDK19 21]

I think the usual method is for the client and server to sync a agreed "general trajectory" [say its a body being blown up] or "flight path"

then the client will do all the small actions e.g. the "falling" of the body, that will fall along the synced flight path,

hence why games often bodies are in slightly different positions on different clients

They are not replicated at all... They'll fly completely random depending on every client's CPU type.

[Pyronick 21]

They are not replicated at all... They'll fly completely random depending on every client's CPU type.

Most (nearly all) games don't synchronize trajectory and final location.

We're reaching the point that bandwidth and latencies are good enough to synchronize trajectory, final location and final pose.

Only the software hasn't been written yet, most middleware physics packages are still fully client-sided. Those that are synchronized usually are modified physics engines. (Source, Second Life, etc)

The progression of physics engines hasn't been that great the last years. nVidia has a monopoly on GPGPU accelerated physics, Intel owns Havok and cooperates with AMD but a competing physics engine hasn't been released yet.

Developers don't want nVidia PhysX middleware because it's nVidia hardware exclusive.

[CarlGustaffa 4]

And yet I'm experiencing desyncs when damage values are transferred to clients :)

Meaning, I don't think we're ready for that just yet.