# Practical real world simulation using PhysX

Hello,

I wish to know to what level of accuracy and precision PhysX can simulate real world scenarios. This is not for games but for more practical and professional application development. Take the following case for instance: a character pushes a chair or box by applying a force on it. I then track the acceleration of the box. How accurate can that acceleration be relative to what it should be really? With that regards how do other physics engines compare? I am aware of the visual effects/performance drawback that occurs when more accuracy and precision is increased. My question is: to what extent can we increase the accuracy and precision? I do not really care about the visual or performance drawback for this question.

Thanks.

Hi,

PhysX simulation is not intended to be quantitatively realistic, and in particular you will find that the friction model is highly simplified. Under some conditions I would expect the results to be close to reality, but under many circumstances such as high velocity impacts, or pendula made with revolute joints, you will find significant deviations from theoretical calculations.

–Mike

This clarifies things quite a bit. Could you give examples of scenarios where PhysX would prove being accurate enough, if possible?

Thanks.

Hello, I’m new to PhysX and didn’t live through the different upgrades. I’m starting with version 4.0 and am bringing this post back to life as I still have some questions on my mind.

I have read that PhysX will be concentrating on the precision of its simulations. Is there any documentation which shows the quantitative precision of PhysX ? or a comparison with some other “real world” simulations which shows that the physX solution is 1-5 or 10 %away from the “real” solution in a best/worst case scenario ? For example in a set of gears or a chain of multi jointed objects, up to what point would the forces calculated by PhysX be representative ?

I got that with “fast” moving actors interacting with “slow” actors and a 10x magnitude difference between actor masses would cause the solver some problems or to diverge from reality. Is that inherant to the fact that PhysX computes stuff with single precision floats ? Are there simplifications in the physical models which ignore some higher order components ? A combination of other optimizations ?

I see throughout this forum that people want to use PhysX for many other things than games. I think it’s awesome ! I too have a project to use PhysX as some sort of real world mechanical solver. Hence my questions on precision.

I have read here : https://devtalk.nvidia.com/default/topic/1046222/physics-modeling/formulation-of-the-temporal-gauss-seidel-tgs-solver/ that a technical doc will come out in a near future. Any update on that ?

Thanks a lot !