Precission of the simulation

[lepalom]

Hi,

I'm new to this library and I just only have run some example. I have found that the dynamics is not precise in the way that for example in the ColladaDemo, I can move a block and put it in a inestable position, not real.

My question is how real is the physics in the library and if it could be modified? (I supossed more real, more slow, less real more fast)

Also in the penetration depth ...

can I trust in the library for a scientific simulation?

Regards,

Leo

[Erwin Coumans]

Hi,
I'm new to this library and I just only have run some example. I have found that the dynamics is not precise in the way that for example in the ColladaDemo, I can move a block and put it in a inestable position, not real.

Instable position, does it mean the object is deactivated too early? You can disable deactivation/sleeping per object, or entirely. In the ColladaDemo, you can press 'd' to disable deactivation. Currently, the demos run fixed framerate, on 60 hertz, and assume OpenGL with vsync turned on.

My question is how real is the physics in the library and if it could be modified? (I supossed more real, more slow, less real more fast)

You can decrease the time step, increase the number of substeps, increate the number of solver iterations to improve simulation accuracy.

Also in the penetration depth ...
can I trust in the library for a scientific simulation?

Only under certain conditions, this library is very accurate. If you can use small timesteps, and enough collision margin, and not too big objects, then the results are very realistic. As long as the penetration between two objects is less then the sum of the collision margins, then the accuracy is very good. With deeper penetrations, the penetration depth is an estimation, mostly useful for games and animation. The time of impact calculations could be used to prevent penetrations in the future but using this feature is experimental, not for scientific simulation.

[lepalom]

Hi,
I'm new to this library and I just only have run some example. I have found that the dynamics is not precise in the way that for example in the ColladaDemo, I can move a block and put it in a inestable position, not real.

Instable position, does it mean the object is deactivated too early? You can disable deactivation/sleeping per object, or entirely. In the ColladaDemo, you can press 'd' to disable deactivation. Currently, the demos run fixed framerate, on 60 hertz, and assume OpenGL with vsync turned on.

Ok, I have found it, now it works better. Thank's

My question is how real is the physics in the library and if it could be modified? (I supossed more real, more slow, less real more fast)

You can decrease the time step, increase the number of substeps, increate the number of solver iterations to improve simulation accuracy.

Ok, I think that I did the question in a wrong way. I understand that if I increase all that things , the result will be better. The question is how about a simulation could be accurated or no.

Also in the penetration depth ...
can I trust in the library for a scientific simulation?

Only under certain conditions, this library is very accurate. If you can use small timesteps, and enough collision margin, and not too big objects, then the results are very realistic. As long as the penetration between two objects is less then the sum of the collision margins, then the accuracy is very good. With deeper penetrations, the penetration depth is an estimation, mostly useful for games and animation. The time of impact calculations could be used to prevent penetrations in the future but using this feature is experimental, not for scientific simulation.

Ok, so, do you think that if I modify this parameters and create this conditions I could have good accurated results in the way that I can simulate in a scientific way a mechanical structure, or robot, etc?

Regards,

Leo