Real-Time Physics Simulation Forum

Re: Numerical Method or Methods used by Bullet

2013-04-15T02:03:19+00:00

The paper is included in Erin Catto's GDC 2005 presentation, which may be found at:
http://code.google.com/p/box2d/downloads/list

The GDC 2013 physics tutorial slides also seem to have been posted there.

Statistics: Posted by rtrius — Mon Apr 15, 2013 2:03 am

Re: Numerical Method or Methods used by Bullet

2013-04-12T23:11:11+00:00

I used google cache to get the html version. hope this helps.

http://group.artek-vfx.com/?/file/downl ... U1M2ViYTU=

Statistics: Posted by Dr.Shepherd — Fri Apr 12, 2013 11:11 pm

Re: Numerical Method or Methods used by Bullet

2013-04-12T17:27:50+00:00

Thank you so much rtrius !

For more details, look up "Sequential Impulses", or the paper
"Iterative Dynamics with Temporal Coherence".

I wonder is you have the link to that paper you're telling. I'll be more than thankful is you have it. The ones that I've found are dead links.

Statistics: Posted by JDTR75 — Fri Apr 12, 2013 5:27 pm

Re: Numerical Method or Methods used by Bullet

2013-04-11T02:30:11+00:00

The rigid body simulation uses symplectic Euler, also called semi-implicit Euler.
Constraints use the maximal coordinate formulation. That is, each rigid body
starts with 6 degrees of freedom and each constraint removes 1 or more
degrees of freedom. The constraints are solved using Projected Gauss-Seidel.
(Basically the same as Gauss-Seidel, but the result is clamped during each iteration.)
For more details, look up "Sequential Impulses", or the paper
"Iterative Dynamics with Temporal Coherence".

Statistics: Posted by rtrius — Thu Apr 11, 2013 2:30 am

Numerical Method or Methods used by Bullet

2013-04-09T01:14:40+00:00

Hi everyone,

Before creating this topic, I was trying to get my answer by searching other posts in the forum but didn't find anything concrete. As the subject says, I would like to know, what are the numerical methods used by Bullet? (searching through the forum I found some other posts that talk about symplectic Euler, Velocity Verlet, Leapfrog integration, etc). I want to know this because I'm doing simulations of different control systems (e.g: simple pendulum, double pendulum, inverted pendulum, etc) and I want to compare the data from Bullet against the differential equations of those systems using MATLAB or Octave for example.

To summarize, I'm gonna make an evaluation on how well Bullet performs the calculations. Of course I'll take in account the computational limitations such as floating point, numerical erros (watch the difference between the data from Bullet and the differential equation resolution), etc.

Regards,
JD

Statistics: Posted by JDTR75 — Tue Apr 09, 2013 1:14 am