Page 1 of 1

Issue with Physics engines.

Posted: 10 Dec 2019, 14:30
by martin
Hello,

I have an issue with the physics engines. For simulation, I was using for long time the default VREP configuration: Bullet 2.78, 50ms time step, etc. for my experiments. Under my point of view, it has always work fine. I am using the Python legacy remote API and working in synchronous mode.
However, in my last experiments, I could see an strange behavior on the physics of the simulation related with the joint. This behavior can be seen here:
https://www.dropbox.com/s/l5rmv0op0qxd2 ... 1.mp4?dl=0

The blue part of the prismatic joint is misaligned with the orange one and a strange behavior is observed. Looking for information, I presume that it is due to physics parameters of the Bullet: Normal CFM, StopCFM y StopERP. However, none value change of these parameters has helped me to solve the problem and become the behavior of the joint similar to a rigid joint.

I was able to reach this changing the physics engine: from Bullet to ODE. And this problem was solved.

However, with this solution, other problem has emerged.
While using Bullet I was able to make my program completely repeatable all times that I run it (same experiment, same seed, all the same). However, I cannot do that with the default configuration of the ODE. When I run the same experiment several times, I get different results each time.

In fact, running the same program and scene, I get a repeatable results with Bullet, but different ones each time I run ODE.

Then my point is:
¿Have you got any idea about how can I do the joints in the Bullet simulation "rigid" or how can I achieve the same repeatability with ODE that I have with Bullet?

Regards!

Re: Issue with Physics engines.

Posted: 11 Dec 2019, 21:22
by fferri
Make sure to respect all the design considerations from Designing Dynamic Simulations.

Re: Issue with Physics engines.

Posted: 12 Dec 2019, 18:55
by martin
Hi fferri,

I think that I've build the scene following those advice. Here, I have a link with the scene and my python code for testing the strange behavior. I think that both are quite simple. The program is "2019_12_12_walker.py""

https://www.dropbox.com/sh/yivivlwjjito ... HueGa?dl=0

Regards!

Re: Issue with Physics engines.

Posted: 16 Dec 2019, 17:20
by martin
Has anyone got any suggestion?