Predictible Physx

[OakWind]

I've ran a few tests and it looks like Physx objects are simi-random in their movement. Is there a way to make them reliably repeatable without scripting a new physx system?

I was dropping a mesh onto a flat surface.

[Tom Havlik]

No.

In fact, realtime physics are very, very inaccurate compared to precomputed and offline calculations. You're resolving thousands of collisions in real time, along with running the rest of the game. It's approximation on top of approximation, and after so long you'll get random results.

Due to the inaccurate nature of physics simulators, you may also notice that every object is slightly "spongy" in behavior. Stuff like stacking boxes may be slightly challenging, or they may bounce on top of each other a couple times when the level loads, until all of the collisions are resolved.

So no, this is not possible, but you'll see that any time that physics accuracy is critical that developers have a "margin of error" or use scripted events / keyframed animations to do this stuff (though this is very seldom in modern games.)

[OakWind]

Well that makes it impossible to make a physics game.

[Elude]

Well that makes it impossible to make a physics game.

Not necessarily, halflife 2 and little big planet both are heavily based on physics for gameplay purposes and even these randomize the physics within the game each play through.

[ambershee]

Indeed. If this were the stumbling block, we'd have never seen any 'physics games'.

[Tom Havlik]

Well that makes it impossible to make a physics game.

. . .but you'll see that any time that physics accuracy is critical that developers have a "margin of error". . .

Read what I say

[OakWind]

Well for my purposes it makes it alot more difficult.

I'm defining a physics based game as: adding objects to a level then starting it to try to achieve a goal of some sort. With no player interaction on the level when the physics are started.

Think :armadillo run, bridge builder, crayon physics(with no interaction when level starts).

I think I'll try to constrain the fps and see if that will increase accuracy. But it looks like I'll have to just deal with the randomness in the design.