2

Let's say I throw a cube and it falls on the ground with 45, 45, 0 rotations (on it's corner). Now in a 'perfect' world, the cube wouldn't consist of atoms, and it would be 'perfect', there would be no wind (or any lesser movement of air) etc. And in the end the cube would stay on it's corner. But we don't live in such a boring 'perfect' world, and the physics emulators should take this into account and they do quite nicely. So the cube falls on it's side.

Now my question is, how random is that? Does the cube always fall on it's left side? Or maybe it depends on Math.random()? Or maybe it depends on current time? Or maybe it depends on some custom random function, that takes not time, but parameters of objects on stage, as it's seed?

Why I am making this question is, that if the randomness wasn't based on time, I probably could cache results of collisions (when objects stop) for their particular initial position to optimize my animation? If I cached the whole animation, I wouldn't care, but If I only cached the end result, I could be surprised that two exactly same situations can evaluate to different results and then the other wouldn't fit my cached version.

I could just check the source for Math.random functions, but that would be a shallow method, as the code is surely optimized, and as such sophisticated randomization isn't needed there, I personally would use something like fallLeft = time % 2. Also, the code could change with time.

Couldn't find anything about AwayPhysics here, so probably it's something new for everyone - that's why I added the parentheses part; the world won't explode if I'll assume one thing and it happens that in AwayPhysics it's opposite, just what's the standard?

Peter O.
  • 32,158
  • 14
  • 82
  • 96
Markus von Broady
  • 313
  • 2
  • 16

1 Answers1

4

I, personally, don't use pre-made physics engines. Instead, when I want one, I write it myself, so I know how they work inside. The reason that the cube tips over is because the physics engine is inaccurate. It can only approximate things like trig functions, square-roots, integrals, et cetera, so instead it estimates them to a few digits of accuracy (15 in Javascript). If you have the case of, say, two perfect circles stacked on top of each other, the angle between them (pi/2) would slowly change to some seemingly random value based on the way the program was approximating pi. Eventually, this tiny error would grow as the circles rolled off of each other, and the top one would just fall. So, in answer to your question, the cube should fall the same way each time if thrown in the same way, but the direction in which it always fell would be effectively random.

mindoftea
  • 816
  • 6
  • 16
  • I don't understand one thing, why the angle changes because of the approximation? Math.floor(x) == Math.floor(x). – Markus von Broady Jul 01 '12 at 10:42
  • Well, imagine the angle is in radians, so 45° straight up would be written as π/4. The actual number π/4 has an infinite number of digits, and so can't be written or used by a computer. Instead of the actual number (0.7853981634...), the computer just uses an approximation (like 0.7854). This is close, but 0.7854 is a little bit greater than 0.7853981634..., so the cube is just past being at 45° and it falls over. – mindoftea Jul 01 '12 at 13:52
  • Oh, so you state that it is deterministic, as same system will always cut the number in same place, that is, the 45 degree will always evaluate to e.g. 0.7853981634. It is perfectly valid point! Unfortunately on Away3D forum admin said AwayPhysics (BulletPhysics) probably isn't deterministic, as they are working to make the framework less random. In the end it all depends on the library, I suppose... thanks for Your answer. – Markus von Broady Jul 01 '12 at 18:21
  • Interesting, I wouldn't have expected that. My guess as to why would be that the accuracy of the approximation approximation is based on the amount of time available for the calculation, so if the CPU feels "behind", it will use a faster, less accurate model? Then the result of the simulation would depend on when it was started, the condition of the processor, and what else the computer is doing, making the result very random, perhaps? – mindoftea Jul 01 '12 at 18:41
  • My guess is that an approximated value can still be a 'conflict', that is, forces pointing in opposing directions will equal. Then some randomization has to be done. It can be a standard random, based on time, a table of precomputed values, that is being repeated over and over or a seed consisting of all objects on scene. In 2nd case results would depend on animation's place in sequence, in 3rd on seemingly meaningless factors, like light vectors. But that's just my lame guesses. – Markus von Broady Jul 02 '12 at 21:07