About real-time simulation

[Page]

Hi,all

I found that different simulation scenarios need to adjust different parameters, such as: 

SetDefaultSuggestedEnvelope(0.03);

SetDefaultSuggestedMargin(0.001),etc.

 It takes me a long time to adjust the parameters each time I simulate different scenarios. Is there a common setting that works for most scenarios?

My model has many Faces, in order to improve efficiency, I used multicore module, I found that the model of hundreds of thousands of faces reached real-time with multi-threading, but the face simulation of 10 million did not reached real-time, I want to know how to simulate the model of millions of face quickly, and how to set the fastest simulation. Choose what solution type, solution model, and related Settings for collision detection.

Best regards,

Page

[Dario Mangoni]

Hi Page,
the answer about collision parameters has a very simple answer: no, there are no one-fits-all parameters.

If you understand what they are meant for (if not, please refer to docs) you can easily understand that they depends on size and speed of the objects at least.
The default values are already set to meaningful values, but that's all.
Anyway, usually in few iterations one can spot the trade-off that is best for him.

About best settings: it's difficult to say without knowing your specific system.
In general it's better to use NSC formulation to allow larger timesteps, but this makes sense only if the timestep is not already limited by finite elements or other stiff objects.
If using NSC then you can pick either PSOR/BB/APGD, but if you have finite elements you need (the more experimental) ADMM+MKL.

For collision detection settings I'm not an expert of Multicore collision, so others might give their opinion.

Anyway, I would strongly encourage you to ask yourself: are all those millions of faces really needed? Can't the mesh be simplified, at least for collision detection?
Even with the best hardware I don't think you will ever be able to reach real-time with millions of faces.

Dario

[Page]

Hi Dario,         

Thank you for your feedback. I am currently using multi-core modules to test collisions, but there is a strange phenomenon. Do you have any feedback? Here is the code and video

Page

[Dario Mangoni]

Hi Page,
I'm sorry but in the latest months we changed the Chrono API quite a bit, so I cannot easily compile your code that is referring to an old version. I would recommend you to recompile Chrono with latest version if you can.

While you others' answers, I can just broadly suggest to:

• reduce stepsize to 1e-3 just to be sure
• check parameters of demo_MCORE_collision_object to see if yours make sense
• make sure inertia/masses properties are meaningful and that there are no extreme values
• your cube explodes even before touching ground: turn on the contact normals drawing so to see if a collision is happening
  
• turn on the collision shapes drawing to double check if you placed them correctly
• put some clear and easy interia properties to body without complex computations to see if they make sense

[Page]

Hi Dario,

Thank you for your suggestion. My project is working normally now. I am currently collecting collision detection information between objects. When collisions occur between rigid bodies,  do not provide any dynamic feedback and effect, only obtain their contact point information, etc. I am using ReportContactCallback or NarrowphaseCallback functions. Do you have any better methods? Looking forward to your reply.